智能系统学报2016，Vol.11Issue(1)：117-123,7.DOI:10.11992/tis.201505037
流形排序的交互式图像分割
Interactive image segmentation based on manifold ranking
摘要
Abstract
The unsatisfactory results of salient object detection have made object segmentation difficult. In this stud⁃y, we obtain accurate object segmentation by combining effective interactive information under the guidance of an initial saliency map. This method obtains the original saliency map using priori boundaries of the manifold ranking of the superpixels, and based on the obtained saliency map, marks the object labels of the non⁃salient object parts and the background labels of the salient background parts. Next, the superpixels for generating the new saliency map are re⁃sorted by manifold ranking with label information, and the newly obtained and original saliency maps are merged. Finally, object segmentation is achieved by adopting the adaptive threshold method for the merged saliency map. Image segmentation experiments with images from the Berkeley segmentation dataset ( BSD) image database demonstrate that the proposed method can correctly segment objects from images and that the true positive rate ( TPR) and the false positive rate ( FPR) results are better than those achieved using the classical maximal similar⁃ity⁃based region merging ( MSRM) image segmentation algorithm.关键词
流形排序/交互式图像分割/显著目标/目标检测/显著图/背景标记/目标标记Key words
manifold ranking/interactive image segmentation/saliency object/object detection/saliency map/background label/object label分类
信息技术与安全科学引用本文复制引用
曾宪华,易荣辉,何姗姗..流形排序的交互式图像分割[J].智能系统学报,2016,11(1):117-123,7.基金项目
国家自然科学基金重点资助项目（ U1401252）；国家自然科学基金资助项目（61075019，61379114）；重庆市基础与前沿研究计划资助项目（ cstc2015jcyjA40036）． （）
